[Cluster-devel] [RFCPATCH] port cluster stable3 to latest corosync API (untested)

[Cluster-devel] [RFCPATCH] port cluster stable3 to latest corosync API (untested)

[Fabio M. Di Nitto]

Hi Ladies and Gents,

this is a preliminary patch to port cman/ccs and cpg users to the new
corosync API. It was mostly a mechanical work based on original Chrissie
patch for cman.

It is _untested_ and on top of:

corosync svn r2125
openais  svn r1847

The changes are mostly int -> size_t and add consts around.

You have to triple check everything I have done because some of the
const char *data get re-casted to something else and then modified
(in some cases I have seen endian conversions happening on the same
const buffer).

Also, we don't see build warning/errors by lack of int -> size_t
conversions. I might have missed some around even if i did try to catch
them all.

Everything builds without warning except some log_printf format
messages.

Let's try to get this sorted by mid next week.

Thanks
Fabio

PS I'll probably test the patch today and if i find issue, post updates
to it.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: coronew-api.diff
Type: text/x-patch
Size: 40495 bytes
Desc: not available
URL: <http://listman.redhat.com/archives/cluster-devel/attachments/20090423/1b9ad622/attachment.bin>

[Cluster-devel] [RFCPATCH] port cluster stable3 to latest corosync API (untested)

[Fabio M. Di Nitto]

On Thu, 2009-04-23 at 13:38 +0200, Fabio M. Di Nitto wrote:
> Hi Ladies and Gents,
> 
> this is a preliminary patch to port cman/ccs and cpg users to the new
> corosync API. It was mostly a mechanical work based on original Chrissie
> patch for cman.
> 
> It is _untested_ and on top of:
> 
> corosync svn r2125
> openais  svn r1847

tested now on top of very latest corosync and openais. The stack starts
without problem and it seems to be running fine.

Unless there are objections I'll land this patch the same day corosync
0.96 and openais 0.95 will be released, let it boil there for a day or
two and then release 3.0.0.rc2.

I only found one bug after all, but it could have been there before the
change:

2 nodes setup + qdisk

service cman start

cluster is formed, daemons are running etc.

fence_node node1

node1 is rebooted just fine.

node1: service cman start <- the node is killed by cman/corosync and not
allowed to join the cluster for no apparent reason.
Testing with and without qdisk didn't make any difference.

It's not always reproducible.

Fabio

[Cluster-devel] [RFCPATCH] port cluster stable3 to latest corosync API (untested)

[Fabio M. Di Nitto]

this patch is now in git stable3.

only diff from the previous version is removal of a bunch of unrequired
const around (side effect of a bug fix in cpg public API).

Please each subsystem maintainer makes sure to test and port to master.

Thanks
Fabio